Talk by Felix Biessmann, MPI for Biological Cybernetics, Tuebingen, Germany. Given to the Redwood Center for Theoretical Neuroscience at UC Berkeley on March 26, 2010.
Abstract. Canonical Correlation Analysis (CCA) (Hotelling, 1936) is a well established statistical learning technique and offers a generic framework for multivariate and multimodal regression problems. In the past years, CCA has proven useful for the analysis of neuroscientific data, as it allows for finding dependencies between data of different dimensionality or sampling rates. However, if the coupling between the variables of interest is not instantaneous but exhibits some delay, the classical CCA algorithm will fail. We propose some simple extensions of CCA that take into account delays in couplings between variables and validate the algorithms on data from two neuroscientific experiments. The first application are recordings from tree shrew V1. We presented pixel noise stimuli and recorded spikes and local field potentials (Lfps) in different frequency bands. Using CCA between stimuli and neural response we can map receptive fields even with Lfps in frequency bands below 10Hz. The second application are simultaneous recordings of neurophysiological activity and BOLD contrast in V1 of the non-human primate during spontaneous activity. The results confirm recent models of neurovascular coupling mechanisms and extend those by revealing the spatio-temporal neurovascular coupling in V1.